Eta cocycles, relative pairings and the Godbillon-Vey index theorem (Q1934642)

This paper generalizes the Atiyah-Patodi-Singer (APS) index theorem [\textit{M. F. Atiyah, V. K. Patodi} and \textit{I. M. Singer}, Math. Proc. Camb. Philos. Soc. 77, 43--69 (1975; Zbl 0297.58008)] to type~III foliations on compact manifolds with boundary, in the case of suspension foliations (foliated bundles); here, type~III refers to the type of foliation von~Neumann algebra. Recall, that the Atiyah-Patodi-Singer index theorem holds for elliptic operators \(D\) on compact manifolds with boundary, and it involves an additional boundary correction term (with respect to the Atiyah-Singer index theorem on closed manifolds), given by the \(\eta\)-invariant of the induced boundary operator \(D^\partial\). Melrose has given another proof of the APS index theorem involving important new ideas: the boundary components are replaced by cylindrical ends, and b-calculus is used [\textit{R. B. Melrose}, The Atiyah-Patodi-Singer index theorem. Wellesley, MA: A. K. Peters, Ltd. (1993; Zbl 0796.58050)]. Let \((Y,\mathcal{F})\) be a foliated manifold defined by a \(\Gamma\)-covering of a closed manifold, \(\widetilde N\to N\), and an action of \(\Gamma\) on a closed manifold \(T\); i.e., \(Y\) is the quotient space \(\widetilde N\times_\Gamma T\) of \(\widetilde N\times T\) by the diagonal \(\Gamma\)-action, and \(\mathcal{F}\) is induced by the foliation of \(\widetilde N\times T\) by the \(\widetilde N\)-fibers; general foliated bundles on closed manifolds are given in this way. Many of the following ideas are valid for general foliations, but let us consider only the above foliated bundle case for the sake of simplicity. In the foliation versions of the index theorem, one considers a leafwise elliptic operator \(D\) on \(Y\). The study of the index theorem in this exotic setting was begun by \textit{A. Connes} for the case where there is a transverse invariant measure, which is used to produce an index number [Proc. Symp. Pure Math. 38, Part 1, Kingston/Ont. 1980, 521--628 (1982; Zbl 0531.57023)]. But, generically, there is no transverse invariant measure (the type~III case). Then a higher foliation index \(\text{Ind}(D)\) was defined also by \textit{A. Connes} as an element in the \(K\)-theory of the foliation \(C^*\) algebra \(C^*(Y,\mathcal{F})\) [Pitman Res. Notes Math. Ser. 123, 52--144 (1986; Zbl 0647.46054)]. To produce index formulas in this type~III case, one tries to pair \(\text{Ind}(D)\) with some cyclic cocycle \(\tau^c\), defined for the subalgebra \(C_c^\infty(Y,\mathcal{F})\subset C^*(Y,\mathcal{F})\). Indeed one tries to realize this pairing for an intermediate \(C^*\) algebra, \(C_c^\infty(M,\mathcal{F})\subset\mathfrak{U}\subset C^*(Y,\mathcal{F})\), which should be big enough to be holomorphically closed in \(C^*(Y,\mathcal{F})\) and contain a representative of \(\text{Ind}(D)\), and should be small enough so that \(\tau^c\) extends to \(\mathfrak{U}\). Finding such a \(\mathfrak{U}\) can be a difficult task. One may also consider the \(C^*\) algebra \(C(T)\rtimes_{\text{r}}\Gamma\), which is Morita equivalent to \(C^*(Y,\mathcal{F})\), and therefore contains \(\text{Ind}(D)\). Then the cyclic cocycles are taken in the subalgebra \(C(T)\rtimes_{\text{alg}}\Gamma\subset C(T)\rtimes_{\text{r}}\Gamma\), and one is confronted with the task of finding an intermediate subalgebra, \(C(T)\rtimes_{\text{alg}}\Gamma\subset\mathcal{A}\subset C(T)\rtimes_{\text{r}}\Gamma\), with the same properties as the above \(\mathfrak{U}\). Suppose from now on that \(\mathcal{F}\) is of codimension one. Then Connes has proved a higher foliation index theorem for a \(2\)-cyclic cocycle \(\tau_{BT}\) for \(C(T)\rtimes_{\text{alg}}\Gamma\), called the Bott-Thurston cocycle: he defined a \(C^*\) algebra \(\mathcal{A}\) satisfying the above properties with \(\tau_{BT}\), and proved an index formula for the pairing \(\langle\text{Ind}(D),[\tau_{BT}]\rangle\), which involves the Godbillon-Vey class of \(\mathcal{F}\), \(GV\in H^3(Y)\). Another treatment of this index formula was given by \textit{H. Moriyoshi} and \textit{T. Natsume} when \(Y\) is of dimension \(3\) [Pac. J. Math. 172, 483--539 (1996; Zbl 0852.58077)]. They defined a cyclic \(2\)-cocycle \(\tau_{GV}\) for \(C^*(Y,\mathcal{F})\), constructed an holomorphically closed intermediate \(C^*\) algebra, \(C_c^\infty(M,\mathcal{F})\subset\mathfrak{U}\subset C^*(Y,\mathcal{F})\), containing \(\text{Ind}(D)\) and an extension of \(\tau_{GV}\); and proved an index formula stating that \(\langle\text{Ind}(D),[\tau_{GV}]\rangle=\int_Y\omega_{GV}\), where \(\omega_{GV}\) is an explicit closed \(3\)-form representing \(GV\); in particular \(\langle\text{Ind}(D),[\tau_{GV}]\rangle\) equals the Godbillon-Vey invariant \(\langle GV,[Y]\rangle\). This is called the Godbillon-Vey index theorem. The numerical version of the APS index theorem for foliations on \(\partial\)-manifolds with a transverse invariant measure was proved by \textit{M. Ramachandran} [J. Differ. Geom. 38, No. 2, 315--349 (1993; Zbl 0787.58040)], and by \textit{P. Antonini} with the cylindrical treatment [Bull. Sci. Math. 137, No. 2, 140--176 (2013; Zbl 1284.58010)]. \textit{E. Leichtnam} and \textit{P. Piazza} proved a higher index version of the APS index theorem for the case of foliated bundles when the group \(\Gamma\) has polynomial growth [J. Reine Angew. Math. 587, 169--233 (2005; Zbl 1088.58010)]. This condition on \(\Gamma\) implies polynomial growth of the leaves, and therefore the existence of a transverse invariant measure, which excludes the type~III case. Its extension to more general foliated manifolds was proved by \textit{M. E. Zadeh}, but still assuming polynomial growth of the leaves [Bull. Sci. Math. 136, No. 2, 201--238 (2012; Zbl 1257.58014)]. Thus the Godbillon-Vey version of the APS index theorem on \(\partial\)-foliated bundles of type III is completely new, which is the great achievement of the paper under review. Now the suspension is of the form \((X_0,\mathcal{F}_0)\) with \(X_0=\widetilde M\times_\Gamma T\), where \(\widetilde M\to M\) is a \(\Gamma\)-covering of a compact \(\partial\)-manifold, giving rise to an associated foliated manifold with cylindrical ends, \((X,\mathcal{F})\), by using the approach of Melrose. Now the index formula is \(\langle\text{Ind}(D),[\tau_{GV}]\rangle=\int_{X_0}\omega_{GV}-\eta_{GV}\), with terms defined like in the case without boundary, and involving what the authors call the Godbillon-Vey \(\eta\)-invariant, \(\eta_{GV}\), which is a type~III \(\eta\)-invariant. The proof is a new approach to index theory and takes a lot of work. One of the main ideas is the use of relative pairings in \(K\)-theory associated to suitable short exact sequences of algebras, as studied by \textit{M. Lesch, H. Moscovici} and \textit{M. J. Pflaum} [J. K-Theory 3, No. 2, 359--407 (2009; Zbl 1175.46066)]. The authors define a short sequence of the form \(0\to C^*(X,\mathcal{F})\to A^*\to B^*\to 0\), called the Wiener-Hopf extension, which is constructed by using translation invariant operators on the cylinders. Then, besides the index class \(\text{Ind}(D)\in K_*(C^*(X,\mathcal{F}))\), there is a relative index class \(\text{Ind}(D,D^\partial)\in K_*(A^*,B^*)\). Now \(\tau_{GV}\) is defined for the smaller algebra \(J_c(X,\mathcal{F})\) of \(\Gamma\)-equivariant smoothing kernels with \(\Gamma\)-compact support. Thus the authors consider a subsequence \(0\to J_c(X,\mathcal{F})\to A_c\to B_c\to 0\) of the above sequence, made of small subalgebras. They obtain an \(A_c\) cyclic \(2\)-cochain \(\tau^r_{GV}\) from \(\tau_{GV}\) through a regularization à la Melrose, and a \(B_c\) cyclic \(3\)-cocycle \(\sigma_{GV}\) by suspending \(\tau_{GV}\) on the cylinder with Roe's \(1\)-cocycle. This \(\sigma_{GV}\) is called the \(\eta\)-cocycle associated to \(\tau_{GV}\). It is proved that \((\tau^r_{GV},\sigma_{GV})\) is a relative cyclic \(2\)-cocycle for \(A_c\to B_c\), producing a relative class \([(\tau^r_{GV},\sigma_{GV})]\in HC^2(A_c,B_c)\). Then the index formula follows from the equality \(\langle\text{Ind}(D),[\tau_{GV}]\rangle=\langle\text{Ind}(D,D^\partial),[(\tau^r_{GV},\sigma_{GV})]\rangle\). But this equality makes sense only when the objects involved in these pairings are defined for the same short exact sequence of algebras. So, like in the above higher index theorem, the authors have to solve the hard problem of defining an intermediate short exact sequence of Banach algebras, \(0\to\mathfrak{J}\to\mathfrak{U}\to\mathfrak{B}\to0\), sitting between the above sequences, where all classes used in the above pairings are defined.